1. Genome sequencing and assembling
Chitta Baral

2. Basic ideas and limitations
Current lab techniques can sequence small (say 700 base pairs) DNA pieces.
Use restriction enzymes to cut DNA pieces
Sort pieces of different sizes using gel electrophoresis and use the sorting to read them
Mapping and Walking
Sequence one piece, get 700 letters, make a primer that allowed you to read the next 700, and work sequentially down the clone
Estimate for human genome sequencing using this method: 100 years
Shotgun sequencing (introduced by Sanger et al. 1977) for sequencing genomes
Obtain random sequence reads from a genome
Assemble them into contigs on the basis of sequence overlaps
Straightforward for simple genomes (with no or few repeat sequences)
Merge reads containing overlapping sequence
Shotgun sequencing is more challenging for complex (repeat-rich) genomes: two approaches

3. Shotgun sequencing – 2 approaches
Hierarchical shotgun approach
Generating an overlapping set of intermediate-sized (e.g. bacterial artificial chromosomes with 200 KB inserts) clones, and keeping a map of that (it took 2 yrs for mapping e-coli)
Subjecting each of these clones to shotgun sequencing, and using the map to get the whole sequence.
Used in S. cerevisiae (yeast), C. elegans (nematode), A. thaliana (mustard weed) and by the International Human Genome Sequencing Consortium (started in 1990, draft made available in 2000)
Whole-genome shotgun (WGS) approach
Generating sequence reads directly from a whole-genome library
Using computational techniques to reassemble in one step.
Used for Drosophila melanogaster (fruit fly) and by Celera Genomics (formed 1998) for human genome.

4. Sequencing small DNA piecesT C
Use DNA cloning or PCR to make multiple copies.
Put in 4 testtubes marked G, A, T and C
In testtube G use restriction enzymes that cuts at G.
Do the above step for the other testubes.
Use gel electrophoresis separately for the content in each testtube.
The data results in the table on the left.
Reading the table we get G has lengths 1, 7, 12, 13, 19; A has lengths 2, 6, 8, 11, 14,15,16; T has length 4, 5, 9, 18 and C has length 3, 10, 17.
This gives us the sequence.

5. The ARACHNE WGS assembler: outline of assembly algorithm
Input data:
Paired end reads obtained by sequencing both ends of a plasmid of known insert size.
Assumes each base in each read has an associated quality score (say one obtained by PHRED program)
Quality score q corresponds to the probability 10-q/10 that the base is incorrect (40 corresponds to 99.99% accuracy)
Initial step: eliminates terminal regions whose quality is low.
Eliminates reads containing very little high-quality sequence
Eliminates known vector sequences and known contaminants (eg. Sequence from the bacterial host or cloning vector)

6. Cont. Overlap detection and alignment
Create a sorted table of each k-letter subword (k-mer) together with its source (which read) and its position within the read.
Exclude k-mers that occur with extremely high frequency
corresponds to highly repeated sequences;
used to increase the efficiency of the overlap detection process
Identify all instances of read pairs that share one or more overlapping k-mer, and a 3 step process (similar to FASTA) to align the reads effciently
(i) Merge overlapping shared k-mers, (ii) Extend the shared k-mers to alignments, (iii) Refine the alignment by dynamic programming.
Some valid alignments may be missed and some invalid ones may result.

7. ARACHNE: Error correction
Error detection and correction
Generate multiple alignments among overlapping reads
Identify instances where a base is overwhelmingly outvoted by bases aligned to it (taking into account the score quality)
Similarly correct occasional inserts and deletes (mostly due to sequencing errors)

8. ARACHNE: Evaluation of alignments
Assign a penalty score to each aligned pair of overlapping reads
Penalty scores are assigned to each discrepant base, based on the sequence quality score at the base and flanking bases on either side.
Discrepancies in high quality sequences are assigned high penalty, and discrepancies in low quality sequences are penalized less heavily.
The penalty scores for individual discrepancies are combined to yield an overall penalty score for the alignment.
Overlaps incurring too high a penalty are discarded
Likely chimeric reads are also detected and discarded
Reads that contain genomic sequence from two disparate locations are termed chimeric.

9. ARACHNE: paired pairs Identification of paired pairs
Paired reads: reads which are known to be related with respect to orientation and distance.
Searches for instances of two plasmids of similar insert size with sequence overlap occurring at both ends. (together called paired pairs)
These instances are extended by building complexes of such pairs
*********** *********
********** **********
********* ********
Collection of paired pairs are merged together into contigs.

10. ARACHNE: Contig assembly
When repeats are absent: correct assembly can be easily obtained by merging all the overlapping reads.
In presence of repeats, false overlaps may arise between reads derived from different copies of a repeat
ARACHNE identifies potential repeat boundaries and avoids assembling contigs across such boundaries
Potential repeat boundary: a read r can be extended by x and y, but x and y don’t overlap
Merge overlapping read pairs that do not cross a marked repeat boundary.

11. ARACHNE: repeat contigs and supercontigs
Detection of repeat contigs – identified 2 ways
Unusually high depth of coverage
Conflicting links to multiple, distinct, non-overlapping contigs, reflecting the multiple regions that flank the repeat in the genome.
aRb, cRd, eRf … will result in –aR-, -cR-, …
Creation of supercontigs
After marking repeat contigs the unmarked contigs (called unitigs) are assembled.
Use forward-reverse links from reads to order and orient unique contigs into supercontigs

12. ARACHNE: Filling gaps in supercontigs
Layout is a set of contigs each of which is an ordered list of contigs with interleaved gaps: corresponding to 2 kind of regions
Regions marked as repeat contigs (which were omitted in supercontig construction)
Regions for which there are insufficient number of shotgun reads to allow assembly
Fill gap using repeat contigs
For every pair of consecutive contigs with an interleaving gap in a supercontig S, the program tries to find a path of pairwise overlapping contigs that fill the gap.
Forward-reverse links from S guide the construction of the path by identifying contigs likely to fall in the gap.

13. Consensus derivation and postconsensus merger
The layout of overlapping reads is converted into consensus sequence with quality scores.
Done by converting pair-wise alignments of reads into multiple alignments, and deriving the consensus base by weighed voting.